Single-cell RNA-seq data analysis of project HTO_testDataset

Demultiplexing with hashtag oligos

# R Options
options(stringsAsFactors=FALSE)

# Required libraries
library(Seurat) # main
library(ggplot2) # plots
library(patchwork) # combination of plots
library(magrittr) # %>% operator

# Other libraries we use
# Knit: knitr
# Data handling: dplyr, tidyr, purrr
# Tables: kableExtra
# Plots: ggsci

# Source plotting functions
source("R/functions_io.R")
source("R/functions_plotting.R")
source("R/functions_analysis.R")
source("R/functions_util.R")

# Knitr default options
knitr::opts_chunk$set(echo=TRUE, cache=FALSE, message=FALSE, warning=FALSE, fig.width=10)

hto_colours = list()
hto_colours$col_hto_global = ggsci::pal_npg()(3)
hto_colours$col_hto_all = ggsci::pal_npg()(factorial(length(params$hto_names)) + 1)
hto_colours$col_hto_collapsed = ggsci::pal_npg()(length(params$hto_names) + 2)
hto_levels = c("Negative", "Doublet", params$hto_names[length(params$hto_names):1])
names(hto_colours$col_hto_collapsed) = hto_levels

if (!file.exists(params$path_out)) dir.create(params$path_out, recursive=TRUE)

Read input data

In this first section of the report, we read 10X data from the files produced by CellRanger:

• barcodes.tsv.gz: All cell barcodes
  
• features.tsv.gz: (Ensembl) ID, name, and type for each feature
  
• matrix.mtx.gz: Counts for all features

and setup a Seurat object. This object includes different data types in separate assays:

• Gene Expresssion in assay RNA
  
• Antibody Capture in HTO and ADT,
  
• CRISPR Guide Capture in Crispr, and
  
• Custom in Custom.

Note that ‘Antibody Capture’ features can correspond to ‘HTO’ and ‘ADT’ and are distinguished based on provided HTO names.

# Load the dataset with its assays and create a Seurat object; pass hto_names so that HTO will be a separate assay
sc = ReadSparseMatrix(params$path_data, project=params$project_id, row_name_column=1, hto_names=params$hto_names)

# If requested: sample at most n cells
if (!is.null(params$downsample_cells_n)) {
  sampled_barcodes = sample(Seurat::Cells(sc), min(params$downsample_cells_n, length(Seurat::Cells(sc))))
  sc = subset(sc, cells=sampled_barcodes)
}

# Discard cells in other assays which have no HTO counts 
original_assay_names = setdiff(Seurat::Assays(sc), "HTO") 
assay_cells_hto_zero = Seurat::GetAssayData(sc, assay="HTO", slot="counts") %>%
  as.data.frame %>%
  colSums() == 0 
if (sum(assay_cells_hto_zero)>0) {
  sc = subset(sc, cells=which(assay_cells_hto_zero))
  warning("Discarded ", sum(assay_cells_hto_zero), " cells with 0 HTO counts.")
}

# Remember original assay names
original_assay_names = setdiff(Seurat::Assays(sc), "HTO") 

summary = purrr::map_dfr(original_assay_names, function(a) {
  data.frame(Assay=a,
             CellsTotal=length(assay_cells_hto_zero),
             CellsWithHtoReads=sum(!assay_cells_hto_zero),
             CellsWithoutHtoReads=sum(assay_cells_hto_zero))
  })

knitr::kable(summary, align="l", caption="Dataset summary") %>% 
  kableExtra::kable_styling(bootstrap_options=c("striped", "hover"), full_width=FALSE, position="left")

Dataset summary

Assay	CellsTotal	CellsWithHtoReads	CellsWithoutHtoReads
RNA	16916	16916	0

Demutliplexing with hashtag oligos (HTOs)

This section of the report shows how cells are assigned to their sample-of-origin.

Normalisation of HTO counts

We start the analysis by normalising raw HTO counts. HTO counts for each cell are divided by the total counts for that cell and multiplied by 10,000. This is then natural-log transformed.

sc = Seurat::NormalizeData(sc, assay="HTO", normalization.method="LogNormalize")

# alternative: CLR
# sc = NormalizeData(sc, assay="HTO", normalization.method="CLR") 

Classification of cells based on normalised HTO data

We assign cells to their sample-of-origin, annotate negative cells that cannot be assigned to any sample, and doublet cells that are assigned to two samples.

# Demultiplex HTOs
sc = Seurat::HTODemux(sc, assay="HTO", positive.quantile=0.95)
  
# Sort Idents levels for nicer plotting
Seurat::Idents(sc) = factor(Seurat::Idents(sc), levels=hto_levels)
sc$hash.ID = factor(sc$hash.ID, levels=hto_levels)
  
# HTO classification results
hash_ID_table = sc@meta.data %>% 
  dplyr::count(hash.ID) %>% 
  dplyr::rename(HTO=hash.ID) %>% 
  dplyr::mutate(Perc=round(n/sum(n)*100,2)) %>% 
  as.data.frame

p1 = ggplot(sc@meta.data %>% dplyr::count(HTO_classification.global), 
            aes(x="", y=n, fill=HTO_classification.global)) + 
  geom_bar(width=1, stat="identity") + 
  coord_polar("y", start=0) + 
  AddStyle(title="HTO global classification results", 
           fill=hto_colours$col_hto_global, 
           xlab="", ylab="")

p2 = ggplot(sc@meta.data %>% dplyr::count(hash.ID), 
            aes(x="", y=n, fill=hash.ID)) + 
  geom_bar(width=1, stat="identity") + 
  coord_polar("y", start=0) + 
  AddStyle(title="HTO classification results", 
           fill=hto_colours$col_hto_collapsed,
           xlab="", ylab="")

p = p1 + p2 + 
  patchwork::plot_annotation(title="HTO classification results") + 
  gridExtra::tableGrob(hash_ID_table, rows=NULL)
p

Visualisation of raw and normalised HTO data

This section of the report visualises raw and normalised HTO data to understand whether the demultiplexing step has worked well.

# Distribution of HTO counts before and after normalisation
hto_t_raw = Seurat::GetAssayData(sc, assay="HTO", slot="counts") %>% 
  as.data.frame %>% t %>% as.data.frame
hto_t_raw_pseudo = hto_t_raw + 1
hto_t_norm = Seurat::GetAssayData(sc, assay="HTO", slot="data") %>% 
  as.data.frame %>% t %>% as.data.frame

p1 = ggplot(hto_t_raw_pseudo %>% tidyr::pivot_longer(tidyr::everything()), aes(x=name, y=value, fill=name)) + 
  geom_violin() + 
  scale_y_continuous(trans="log2") + 
  AddStyle(title="HTO raw counts", 
           fill=hto_colours$col_hto_collapsed, 
           legend_title="HTO", 
           xlab="", ylab="")
  
p2 = ggplot(hto_t_norm %>% tidyr::pivot_longer(tidyr::everything()), 
            aes(x=name, y=value, fill=name)) + 
  geom_violin() + 
  AddStyle(title="HTO normalised counts", 
           fill=hto_colours$col_hto_collapsed, 
           legend_title="HTO",
           xlab="", ylab="")

p = p1 + p2 & theme(legend.position="bottom") 
p = p + patchwork::plot_annotation("HTO counts before and after normalisation") + 
  patchwork::plot_layout(guides = "collect")
p

Pairs of raw (top) and normalised (bottom) HTO counts are visualised to confirm mutal exclusivity in singlet cells. Data points correspond to measured HTO counts per HTO, colours correspond to the assigned samples-of-origin.

n = DfAllColumnCombinations(x=hto_t_raw_pseudo, cell_classification=sc$hash.ID)

# plot
p = ggplot(n, aes(x=value1, y=value2, color=cell_classification)) + 
  geom_point() + 
  scale_x_continuous(trans="log2") + scale_y_continuous(trans="log2") + 
  AddStyle(col=hto_colours$col_hto_collapsed)

p = p + facet_grid(name2~name1, drop=FALSE) + 
  theme(axis.title.x=element_blank(), strip.text.x=element_text(size=10, color="black"),
        axis.title.y=element_blank(), strip.text.y=element_text(size=10, color="black"),
        strip.background = element_rect(colour="white", fill="lightgrey"),
        legend.position="bottom", 
        axis.text.x=element_text(angle=45, hjust=1, vjust=0.5)) + 
  patchwork::plot_annotation("Raw HTO counts")
p

n = DfAllColumnCombinations(x=hto_t_norm, cell_classification=sc$hash.ID)
p = ggplot(n, aes(x=value1, y=value2, color=cell_classification)) + 
  geom_point() + 
  scale_x_continuous(trans="log2") + scale_y_continuous(trans="log2") + 
  AddStyle(col=hto_colours$col_hto_collapsed)

p = p + facet_grid(name2~name1, drop=FALSE) + 
  theme(axis.title.x=element_blank(), strip.text.x=element_text(size=10, color="black"),
        axis.title.y=element_blank(), strip.text.y=element_text(size=10, color="black"),
        strip.background = element_rect(colour="white", fill="lightgrey"),
        legend.position="bottom") + 
  patchwork::plot_annotation("Normalised HTO data")
p

The following ridge plots visualise the enrichment of assigned sample-of-origin for the respective normalised HTO counts.

# Group cells based on HTO classification 
p_list = RidgePlot(sc, assay="HTO", features=rownames(Seurat::GetAssay(sc, assay="HTO")), 
                   same.y.lims=TRUE, cols=hto_colours$col_hto_collapsed, combine=FALSE)
for (i in seq(p_list)) p_list[[i]] = p_list[[i]] + AddStyle(legend_title="Classified cells")
p = patchwork::wrap_plots(p_list, ncol = 2) + 
  patchwork::plot_annotation("Normalised HTO data") + 
  patchwork::plot_layout(guides = "collect") & 
  theme(legend.position="bottom") 
p

Lastly, we compare the number of features between classified cells.

# Number of features in the different cells
nfeature_metrics = grep("_HTO", 
                        grep("nFeature_", colnames(sc[[]]), v=TRUE),
                        v=TRUE,
                        invert=TRUE)
p = VlnPlot(sc, features=nfeature_metrics, idents=levels(Seurat::Idents(sc)), pt.size=0) + 
  geom_violin(color=NA) +
  AddStyle(title="Number of features for HTO-classified cells", 
           fill=hto_colours$col_hto_collapsed, 
           legend_title="Classified cells",
           legend_position="bottom", 
           xlab="")
p

Remove cells classified as doublet or negative

This section of the report states the number of cells that remain after negative and doublet cells are removed.

sc.all = sc
sc = subset(sc, idents=c("Negative", "Doublet"), invert=TRUE)
sc

## An object of class Seurat 
## 18984 features across 13843 samples within 2 assays 
## Active assay: RNA (18976 features, 0 variable features)
##  1 other assay present: HTO

Preliminary visualisation of demultiplexed RNA data

This section of the report provides first insights into your RNA dataset based on a preliminary pre-processing of the RNA data using the standard scRNA-seq workflow.

sc.all = prelim.analysis(sc=sc.all, mt=params$mt_names, pc_n=10)
sc = prelim.analysis(sc=sc, mt=params$mt_names, pc_n=10)

Visualisation with UMAP

We use a UMAP to visualise and explore a dataset. The goal is to place similar cells together in 2D space, and learn about the biology underlying the data. Cells are color-coded according to the assigned sample-of-origin.

Take care not to mis-read a UMAP:

• Parameters influence the plot (we use defaults here)
  
• Cluster sizes relative to each other mean nothing, since the method has a local notion of distance
  
• Distances between clusters might not mean anything
  
• You may need more than one plot

For a nice read to intuitively understand UMAP, see https://pair-code.github.io/understanding-umap/.

Before filtering

p = Seurat::DimPlot(sc.all, reduction="umap", group.by="hash.ID", cols=hto_colours$col_hto_collapsed) + 
  AddStyle(title="UMAP, cells coloured by HTO classification, including doublets and negatives", 
           legend_title="Cell classification", 
           legend_position="bottom")
p

After filtering

# Plot UMAP after HTO filtering
p = Seurat::DimPlot(sc, reduction="umap", group.by="hash.ID", cols=hto_colours$col_hto_collapsed) + 
  AddStyle(title="UMAP, cells coloured by HTO classification, singlets only", 
           legend_title="Cell classification", 
           legend_position="bottom")
p

Write out demultiplexed data

Finally, demultiplexed RNA data are written back to file.

# Save each sample in a separate directory
samples = levels(Seurat::Idents(sc))
demux_samples_paths = c()
for (s in samples) {
  p = ExportSeuratAssayData(sc[,Seurat::Idents(sc)==s], 
                            dir=file.path(params$path_out, s), 
                            assays=original_assay_names, 
                            slot="counts",
                            include_cell_metadata_cols=c("HTO_maxID", 
                                                         "HTO_classification", 
                                                         "HTO_classification.global", 
                                                         "hash.ID"))
  demux_samples_paths = c(demux_samples_paths, p)
}

message(demux_samples_paths)

Parameter table

The following parameters were used to run the workflow.

out = scrnaseq_params_info(params=params)

knitr::kable(out, align="l") %>% 
  kableExtra::kable_styling(bootstrap_options = c("striped", "hover"), full_width=FALSE, position="left")

Name	Value
project_id	HTO_testDataset
path_data	test_datasets/10x_pbmc_hto_GSE108313/counts
path_out	test_datasets/10x_pbmc_hto_GSE108313/demultiplexed
hto_names	htoA=htoA, htoB=htoB, htoC=htoC, htoD=htoD, htoE=htoE, htoF=htoF, htoG=htoG, htoH=htoH
mt_names	^MT-
col	palevioletred
downsample_cells_n

Software versions

This report was generated using the scrnaseq GitHub repository. Software versions were collected at run time.

out = scrnaseq_session_info()

knitr::kable(out, align="l") %>% 
  kableExtra::kable_styling(bootstrap_options = c("striped", "hover"), full_width=FALSE, position="left")

Name	Version
ktrns/scrnaseq	94536c83158fd9a9cc989fc27639fcd95d300385
R	R version 3.6.1 (2019-07-05)
Platform	x86_64-apple-darwin15.6.0 (64-bit)
Operating system	macOS Mojave 10.14.6
Packages	ape5.3, assertthat0.2.1, backports1.1.5, bitops1.0-6, caTools1.17.1.3, cli2.0.2, cluster2.1.0, codetools0.2-16, colorspace1.4-1, cowplot1.0.0, crayon1.3.4, data.table1.12.6, digest0.6.25, dplyr0.8.3, evaluate0.14, fansi0.4.0, farver2.0.1, fitdistrplus1.0-14, future1.15.1, future.apply1.3.0, gdata2.18.0, ggplot23.2.1, ggrepel0.8.1, ggridges0.5.1, ggsci2.9, globals0.12.5, glue1.4.1, gplots3.0.1.1, gridExtra2.3, gtable0.3.0, gtools3.8.1, highr0.8, hms0.5.2, htmltools0.4.0, htmlwidgets1.5.1, httr1.4.1, ica1.0-2, igraph1.2.4.2, irlba2.3.3, jsonlite1.6.1, kableExtra1.1.0, KernSmooth2.23-16, knitr1.26, labeling0.3, lattice0.20-38, lazyeval0.2.2, leiden0.3.1, lifecycle0.1.0, listenv0.8.0, lmtest0.9-37, lsei1.2-0, magrittr1.5, MASS7.3-51.4, Matrix1.2-18, munsell0.5.0, nlme3.1-142, npsurv0.4-0, patchwork1.0.0, pbapply1.4-2, pillar1.4.2, pkgconfig2.0.3, plotly4.9.1, plyr1.8.4, png0.1-7, purrr0.3.3, R.methodsS31.7.1, R.oo1.23.0, R.utils2.9.2, R62.4.1, RANN2.6.1, RColorBrewer1.1-2, Rcpp1.0.3, RcppAnnoy0.0.14, RcppParallel4.4.4, readr1.3.1, reshape21.4.3, reticulate1.13, rlang0.4.6, rmarkdown1.18, ROCR1.0-7, RSpectra0.16-0, rstudioapi0.11, rsvd1.0.2, Rtsne0.15, rvest0.3.5, scales1.1.0, sctransform0.2.0, sessioninfo1.1.1, Seurat3.1.5, stringi1.4.3, stringr1.4.0, survival3.1-8, tibble2.1.3, tidyr1.0.0, tidyselect0.2.5, tsne0.1-3, uwot0.1.5, vctrs0.2.0, viridisLite0.3.0, webshot0.5.2, withr2.1.2, xfun0.11, xml21.2.2, yaml2.2.0, zeallot0.1.0, zoo1.8-6